Machine Learning Model Management with MLflow

This project demonstrates an end-to-end Machine Learning Operations (MLOps) workflow using MLflow for experiment tracking, model management, artifact generation, and model registration.

The repository provides a modular environment for training multiple machine learning models, logging experiments, comparing their performance, and managing registered models through the MLflow web interface.

---

Table of Contents

1. Overview
2. Features
3. Technologies
4. Installation
5. Project Contents
6. Author
7. Contact

---

Overview

This project implements a complete workflow for machine learning experiment management using MLflow.

The repository includes scripts for:

• Exploratory Data Analysis (EDA).
• Feature visualization.
• Training multiple machine learning algorithms.
• Logging metrics, parameters, and artifacts.
• Registering trained models.
• Comparing model performance through the MLflow Tracking UI.

The experiments are conducted using the Breast Cancer Wisconsin Dataset, where several supervised classification algorithms are trained and evaluated.

The entire application is containerized using Docker, providing a reproducible environment for experimentation.

---

Features

• 📊 Experiment tracking with MLflow.
• 🧠 Training multiple machine learning models.
• 📦 Automatic artifact generation.
• 🗂️ Model registry support.
• 🐳 Docker-based deployment.
• 🖥️ Compatible with Windows.

---

Technologies

This project uses:

• Python 3.11+
• Pandas
• NumPy
• Matplotlib
• MLflow
• Docker

---

Installation

Prerequisites

• Python 3.11+
• Docker Desktop
• Internet connection

1. Clone the repository

git clone https://github.com/fabriciolopretto/Administracion-de-Modelos.git
cd Administracion-de-Modelos

2. Start Docker Desktop

Ensure Docker Desktop is running before building the container.

3. Build and run the Docker container

From the directory containing the Dockerfile, execute:

docker build -t image_name .

docker run -it \
  --name container_name \
  -p 5000:5000 \
  -v "$(pwd)/TP_Final/mlflow/experiments/models/mlruns:/app/mlruns" \
  image_name

4. Execute the project scripts

Data Analysis

python Distributions.py
python HeatMap.py

Model Training

python RegLog.py
python KNN.py
python SVC.py
python TreeClasf.py

Artifact Generation and Model Registration

python registro_reg_log.py
python predictions_reglog_model.py

5. Open the MLflow UI

Once the container is running, open your browser and navigate to:

http://localhost:5000

---

Project Contents

The repository includes:

• Machine learning training scripts.
• Exploratory data analysis notebooks and scripts.
• MLflow experiment logs.
• Registered model artifacts.
• Dockerfile for reproducible deployment.
• requirements.txt
• config.ini for PostgreSQL configuration.
• MLflow run history and artifacts.

---

Dataset

The experiments use the Breast Cancer Wisconsin Dataset, which contains numerical features extracted from digitized images of breast cell nuclei. The objective is to classify tumors as benign or malignant.

---

Author

Lic. Fabricio Lopretto

---

Contact

For questions, suggestions, or collaborations, please feel free to get in touch.